Journal of Biomechanics 32 (1999) 443—451 Erratum An error occured on p. 169 of the above article, whereby Fig. 4 was reproduced incorrectly. The complete paper with the correct version of Fig. 4 is reprinted below. The publisher apologises most sincerely to the authors and the readers for any inconvenience caused by this error. Tissue stresses and strain in trabeculae of a canine proximal femur can be quantified from computer reconstructions B. Van Rietbergen , R. Müller, D. Ulrich, P. Rüegsegger, R. Huiskes * Orthopaedic Research Lab, Institute of Orthopaedics, University of Nijmegen, P.O. Box 9101, 6500 HB Nijmegen, The Netherlands Institute for Biomedical Engineering, University of Zu( rich and Swiss Federal Institute of Technology (ETH), Zu( rich, Switzerland Received 29 December 1997; accepted 21 September 1998 Abstract A quantitative assessment of bone tissue stresses and strains is essential for the understanding of failure mechanisms associated with osteoporosis, osteoarthritis, loosening of implants and cell- mediated adaptive bone-remodeling processes. According to Wolff’s trajectorial hypothesis, the trabecular architecture is such that minimal tissue stresses are paired with minimal weight. This paradigm at least suggests that, normally, stresses and strains should be distributed rather evenly over the trabecular architecture. Although bone stresses at the apparent level were determined with finite element analysis (FEA), by assuming it to be continuous, there is no data available on trabecular tissue stresses or strains of bones in situ under physiological loading conditions. The objectives of this project were to supply reasonable estimates of these quantities for the canine femur, to compare trabecular-tissue to apparent stresses, and to test Wolff ’s hypothesis in a quantitative sense. For that purpose, the newly developed method of large-scale micro-FEA was applied in conjunction with micro-CT structural measurements. A three-dimensional high-resolution computer reconstruction of a proximal canine femur was made using a micro-CT scanner. This was converted to a large-scale FE-model with 7.6 million elements, adequately refined to represent individual trabeculae. Using a special-purpose FE-solver, analyses were conducted for three different orthogonal hip-joint loading cases, one of which represented the stance-phase of walking. By superimposing the results, the tissue stress and strain distributions could also be calculated for other force directions. Further analyses of results were concentrated on a trabecular volume of interest (VOI) located in the center of the head. For the stance phase of walking an average tissue principal strain in the VOI of 279 strain was found, with a standard deviation of 212 lstrain. The standard deviation depended not only on the hip-force magnitude, but also on its direction. In more than 95% of the tissue volume the principal stresses and strains were in a range from zero to three times the averages, for all hip-force directions. This indicates that no single load creates even stress or strain distributions in the trabecular architecture. Nevertheless, excessive values occurred at few locations only, and the maximum tissue stress was approximately half the value reported for the tissue fatigue strength. These results thus indicate that trabecular bone tissue has a safety factor of approximately two for hip-joint loads that occur during normal activities. 1999 Elsevier Science Ltd. All rights reserved. Keywords: Trabecular bone; Bone mechanical properties; Computed tomography; Finite element analyses; Bone architecture * Corresponding author. Tel.: 0031 24 361 4476; fax: 0031 24 454 0555; e-mail: r.huiskes@orthp.azn.nl. Original PII: S0021-9290(98)00150-X. Present address: Institute for Biomedical Engineering, University of Zürich and Swiss Federal Institute of Technology (ETH), Zürich, Switzerland. Present address: Orthopedic Biomechanics Laboratory, Harvard Medical School (BIDMC), Boston, USA. 0021-9290/99/$ — see front matter 1999 Elsevier Science Ltd. All rights reserved. PII: S 0 0 2 1 - 9 2 9 0 ( 9 9 ) 0 0 0 2 4 - X 444 B. Van Rietbergen et al. / Journal of Biomechanics 32 (1999) 443—451 1. Introduction The main function of trabecular bone is to distribute mechanical loads from articular surfaces to the diaphyses of the long bones. The load-transfer pathway is largely determined by the internal architecture of the bone, since the individual trabeculae constitute the actual load-carrying structure. According to Wolff ’s trajectorial hypothesis, the trabecular architecture is such that minimal tissue stresses are paired with minimal weight. This paradigm at least suggests that, normally, stresses and strains should be distributed rather evenly over the trabecular architecture. So far, however, there have been no possibilities for a quantitative evaluation of this paradigm. It is not known, for example, if, and to what extent, an even distribution is possible for the actual tissue stresses and strains, or, if this is only possible for the average tissues stresses and strains during a loading cycle. Nor is it known to what extent this expected even distribution of tissue stresses and strains is affected if the bone architecture or the external loads are changed and what the ‘safety factor’ of the bone is for changes in its loading. Such changes in architecture or loading can be due to, for example, osteoporosis or the placement of an implant. Quantitative knowledge about the tissue stress and strain distribution thus could be a key factor to quantify bone integrity according to Wolff ’s hypothesis, but is also essential for the understanding of failure mechanisms associated with osteoporosis, osteoarthritis, loosening of implants and cell-mediated load adaptive bone remodeling processes. So far, however, no methods have been developed that can be used to measure tissue stresses or strains, not even in vitro. As an alternative, methods based on the finite element analysis (FEA) have been used to calculate rather than measure tissue stress and strain conditions. In these studies, however, the bone tissue was considered as a continuum. Such continuum models can only be used to calculate average tissue stresses and strains, and not those in the individual trabeculae. Several authors have attempted to obtain information about trabecular stresses and strains, but only with respect to small bone samples. In an early study, trabecular architecture was represented as a repetitive structure of unit cells (Beaupré and Hayes, 1985). More recent studies have used new techniques that enable the FE-analysis of realistic trabecular architectures in detail (Fyhrie et al., 1992; Hollister et al., 1993; Van Rietbergen et al., 1995). These techniques were based on high-resolution imaging techniques, such as serial sectioning (Odgaard et al., 1994) or micro-CT scanning (Feldkamp et al., 1989; Rüegsegger et al., 1996), in combination with newly developed, iterative FE-solvers (Hollister and Kikuchi, 1993; Van Rietbergen et al., 1995, 1996). With these techniques, the detailed three-dimensional architecture of bone samples can be digitized and converted to large- scale FE-models from which tissue stresses can be calculated. Applying these methods to cubic bone samples, it was found that the actual tissue stresses can be much higher than those calculated from continuum models (Hollister and Kikuchi, 1993; Van Rietbergen et al., 1995). Although these studies have produced valuable information about load transfer in trabecular architectures, this cannot be transferred to the situation of bone in situ, as the boundary forces for an excised specimen are not representative for the intact situation. In other studies, a homogenization sampling procedure, incorporating large-scale FE-models representing small samples of trabecular bone in detail, was applied to obtain the trabecular tissue stresses throughout larger pieces of bones or whole joints (Hollister et al., 1993; Hollister and Goldstein, 1993). With this procedure, however, assumptions about boundary conditions can seriously influence the results as well. To obtain dependable information about trabecular stresses and strains, in short, one cannot escape the necessity of representing the structure and its loading conditions realistically. The objective of this study was to determine physiological in situ stresses and strains in trabecular bone tissue of a proximal femur. For this purpose, a microstructural finite element model was used, that represents a whole proximal femur in detail. The question was asked of how the load is transferred through the trabecular architecture for a variety of hip-joint forces, if an external hipjoint force exists that provides a uniform distribution of tissue stresses and strains as predicted by Wolff’s law; how much variety in trabecular stresses and strains occurs for other force directions; whether the extent of this range can be estimated from average tissue values as determined from apparent-level FE models; and what the ‘safety factor’ of the bone architecture is for hip-joint loads that occur during normal activities. The model used in the present study represents a canine proximal femur. The choice for this model was based on size limitations of present computer-reconstruction methods and FE-solvers, and on the fact that detailed in vivo joint force data are available for the dog. 2. Methods The right femur of a small dog was selected from a stock of six. On micro-radiographs, this femur looked average in shape and trabecular architecture and showed no remains of a growth plate. A body weight of 13 kg was estimated for the dog, based on the size of the femur. A computer reconstruction of the proximal 27 mm of the femur was made using a micro-CT scanner (Scanco Medical, Bassersdorf, Switzerland). Since the dimensions of the proximal femur in the medial-lateral direction exceeded the bore size of the micro-CT scanner (16 mm), B. Van Rietbergen et al. / Journal of Biomechanics 32 (1999) 443—451 a part of the trochanter was cut from the bone before scanning, using a 0.3 mm wire-saw, and was scanned separately. The nominal resolution of the scanner is 14 lm (Rüegsegger et al., 1996), but to reduce the data set and scan time a voxel size of 35 lm in all directions was chosen. With this voxel size, the total scan time was close to 6 h. Before segmentation the voxel size was increased to 70 lm to further reduce the number of voxels. Both three-dimensional reconstructions were merged into a new voxel grid and positioned such that the cut faces were at a 0.3 mm distance. The cut trabeculae and cortical regions on either side of the gap could be identified on the sequential images that form the voxel grid, and the gap was repaired by manually editing each image. The resulting computer reconstruction measured 29.3; 19.3;27.0 mm and was represented by 418;275;385 voxels of which 7.3 million represented bone tissue. The quality of the reconstruction was judged by comparing a simulated micro-radiograph made from the computer reconstruction to a real one made earlier (Fig. 1). The simulated radiograph was produced by summing voxel densities in the anterior-posterior direction, assigning a density of 1 for voxels representing bone tissue and zero for the voxels representing marrow. The summed values were linearly scaled to represent a gray level in a digital image such that white represents the highest value and black represents zero. By comparison of the simulated and real micro-radiograph, it was concluded that the computer reconstruction adequately represented typical features as shown on the real micro-radiograph, although some loss of detail is apparent due to the limited resolution (70 lm) of the computer reconstruction. An artificially created cup was merged with the reconstruction of the femur in order to apply realistic loading 445 conditions later. The cup was modeled separately as a 7 mm thick half-hemisphere built of voxels that had the same sizes and orientations as those used for the bone reconstruction. The inner radius of the cup was chosen approximately one voxel smaller than the radius of the femoral head such that cup and femoral head would overlap when their centers are aligned. The cup was positioned in the bone reconstruction grid such that it covered the anterior—medial—superior quadrant of the femoral head, which is the region in which the resultant hip-joint force acts (Bergmann et al., 1984; Page et al., 1993). After merging both reconstructions the bone—cup interface disappeared at most locations since the reconstruction of the cup and that of the bone slightly overlap. In this way an artificial cup was created that was fixed to the femoral head at most of its inner surface. In some regions, however, (notably the region were the femoral fovea and the round ligament are located) the overlap between the cup and the femoral head reconstruction was too small to bridge the gap, and in these regions a gap between the cup and the femoral head remained. Forces were applied to the cup in the medial, anterior and superior directions. Near the force application points, the thickness of the cup was gradually increased to form a plateau for a smooth distribution of the loads to the femoral head. The three-dimensional computer reconstruction was converted to a large-scale FE-model by simply converting all voxels that represent bone tissue or cup to eightnode brick elements in the FE-model. This resulted in a FE-model with a total of 7.6 million elements and 9.1 million nodes (Fig. 2). The same linear elastic and isotropic material properties, with a Young’s modulus of 15 GPa and a Poisson’s ratio of 0.3 were assigned to Fig. 1. Comparison of a micro-radiograph of the proximal part of the dog femur (left), with a simulated radiograph of the 3-D reconstruction (right). The 3-D reconstruction is exactly the same as the FE-model but without the cup. The simulated radiograph was made by summing element densities in the direction normal to the projection. The gray levels in the plot are linearly scaled with the calculated projected density, such that white corresponds to the highest value and black to the lowest. Note that typical structures seen on the real radiograph can be recognized in the simulated radiograph, although some loss of detail is apparent due to the limited resolution of the computer reconstruction. 446 B. Van Rietbergen et al. / Journal of Biomechanics 32 (1999) 443—451 elements representing trabecular bone, cortical bone and the cup. A special-purpose iterative FE-solver, implementing a ‘Row-By-Row’ matrix—vector multiplication algorithm was used to solve this large FE-problem (Van Rietbergen et al., 1996). Approximately 4 GByte memory and 30 h of CPU-time were needed on the Cray C90 supercomputer which was used for the calculations. The FE-problem was solved three times to represent three load cases out of a daily loading cycle (Bergmann et al., 1984; Page et al., 1993). In the first load case (Fig. 3a), a 40 N force directed laterally was applied to the medial side of the cup, in the second case a 50 N force directed posteriorly was applied to the anterior side, and in the third case a 200 N force directed distally to the superior side of the cup. The last case represents the hip joint force during the stance phase of walking. For all cases, the displacements of nodes at the distal end of the femur were fully constrained in all three spatial directions. Since this is a linear FE-model, any resultant force acting towards the center of the head in the anterior— medial—superior quadrant can be approximated by scaling and superimposing the results of these three analyses. In this way, results were calculated for another 88 load cases representing a resultant force in the anterior—medial— superior quadrant at 10° intervals (Fig. 3). For each element, the superimposed strain vector was calculated from F F F e"e 0 cos cos h#e 0 sin cos h#e 0 sin h, F F F (1) where e is the strain vector for load case i, F the magnitude G 0 of the resultant force and , h, F , F and F as defined in Fig. 3. The magnitude of the resultant force F was 0 linearly interpolated from the three load cases using F "f F #f F #f F 0 (2) with interpolation functions: Fig. 2. The FE-model of the dog’s femur with artificial cup. The model is built of 7.6 million cubic brick elements of 70 lm, in size and has a total of 27.3 million degrees of freedom. In the plot, some of the posterior cross sections are removed to show how the trabecular bone is modeled with the brick elements. f " 1! 90 h 1! , 90 h h f " 1! , f " . 90 90 90 The notion that results for the tissue level stresses and strains can be obtained by superimposing those Fig. 3(a). Simulated radiographs of the FE-model with cup in the AP and ML direction with the forces applied for the three load cases. Also indicated is the VOI in the femoral head. Fig. 3(b). The coordinate system used to describe the direction of the resultant joint force. Two coordinate angles and h and three orthonormal forces F , F and F are used to define the direction and magnitude of the resultant force. The center of the sphere corresponds to the center of the femoral head. Only resultant forces acting in the highlighted quadrant are investigated. B. Van Rietbergen et al. / Journal of Biomechanics 32 (1999) 443—451 calculated from three orthogonal hip-joint forces was based on recent findings that the contact areas in the human hip-joint are located near the periphery of the cup with no contact in the region near the center of the acetabular cup where the femoral fovea and the round ligament inhibit contact (Eisenhart-Rothe et al., 1997; Bay et al., 1997), and on the observation that the anatomy of the pelvis favors load transfer in orthogonal directions rather than in other directions. In the analyses of the data we concentrated on a 7 mm cubic volume of interest (VOI) in the center of the head (Fig. 3a). The number of elements in this volume was 608,463 and the number of nodes was 815,967. From these numbers a finite element fractal dimension of 2.58 was calculated (Van Rietbergen et al., 1996). Since this part has a plate-like architecture, the fractal dimension indicates that, on average, three elements were present in a cross section of the trabeculae. For this VOI, histograms for the absolute value of the maximal principal stress and strain, for the Von Mises equivalent stress and for the Strain Energy Density (SED) distribution were calculated. For each of these distributions, the average 447 values and standard deviations were calculated to quantify a range for the tissue stresses and strains. In the following we will simply write ‘principal stress’ where we mean ‘the largest component (in an absolute sense) of the principal stress’, and similarly for ‘principal strain’. To compare the actual trabecular values determined here with those that can be obtained from continuous, apparent-level FE models, the average values over the VOI were considered. For the 200 N force representing the stance phase of walking, principal strains were calculated as well in a 1.4 mm cubic volume located within the cortical bone close to the periosteal surface, on the medial side of the femoral neck. This volume was chosen since it enables comparison of calculated strains with literature values measured from in vivo strain-gauge measurements in dogs (Page et al., 1993). 3. Results A contour plot of the SED distribution for the load case representing the stance phase of walking (case 3) Fig. 4. Contour plot of the strain energy density distribution for a 200 N load acting on the superior side of the cup, representing the stance phase of loading. Red areas indicate high values whereas in white areas the SED is close to zero. Some of the posterior cross sections are removed to show how the load is transferred through the trabecular bone. Also indicated in this plot is the location of the VOI for which histograms of the tissue stresses and strains are calculated. 448 B. Van Rietbergen et al. / Journal of Biomechanics 32 (1999) 443—451 Fig. 5. Histograms representing the tissue principal stress, principal strain, Von Mises stress and SED distribution in the VOI for the 200 N load representing the stance-phase of walking. The average value, standard deviation and maximum value found for any element in the VOI are indicated as well. demonstrates how the load is transferred from the cup, through the trabecular network to the cortex (Fig. 4). The values and distribution of the SED in the cortical region are very similar to those determined in earlier studies, using continuum models of a canine femur (Weinans et al., 1993). For the VOI, however, it was found that the actual tissue principal stress, principal strain, Von Mises stress and SED value could deviate considerably from the average tissue value (Fig. 5). In the histograms of Fig. 5 the average values, the standard deviations and the maximum values calculated for any element in the VOI are indicated as well. For the principal strain, for example, the average value in the VOI was e "279 lstrain, the standard deviation e "212 4 1" lstrain, and the highest value for any element in the VOI was 3731 lstrain. Similar to normal distributions, however, it was found that for 96% of all tissue material the principal strains were in the range defined by e $2;e , and for 69.2% of the tissue, the values did 4 1" not even exceed the range e $e . The principal strain 4 1" in the medial cortex of the femoral neck was larger than that in the trabecular bone tissue and averaged 958 lstrain. The shape of the tissue stress distribution in the VOI resembles that of the strain distribution. The average tissue stress was 3.88 MPa, the standard deviation 3.04 MPa and the largest value found was 60.2 MPa. The average values of the principal strain in the VOI for all 91 load cases, plotted in the contour plot of Fig. 6a, are largely determined by the magnitude of the resultant force vector. One can think of this plot as a projection of the average principal strain onto a sphere, as a function of the two coordinate angles and h. The values range from 57 lstrain for a 41 N resultant force in the "10, h"0 direction, to 279 lstrain for the 200 N force directed distally at h"90. A plot of the standard deviations as a function of the coordinate angles would look similar, since the magnitude of the standard deviations is linearly related to that of the average value. If, however, the standard deviation is normalized by the average value, the effect of the force magnitude is eliminated and a plot results in which high values correspond to resultant force directions for which a relatively large standard deviation for the trabecular tissue strain distribution in the VOI is found, and low values correspond with force directions for which a more narrow distribution is found. The normalized standard deviations for the principal strain distribution curves ranged from 54.4% of the average tissue strain for a resultant force in the "30°, h"50° direction to 99.0% of the average for a resultant force in the "80°, h"10° direction (Fig. 6b). The high values for forces acting on the anterior side (the yellow area in Fig. 6b) thus indicate that the trabecular architecture is not well adapted to forces acting in the posterior B. Van Rietbergen et al. / Journal of Biomechanics 32 (1999) 443—451 449 Fig. 6. (a) Contour plot of the average tissue principal strain in the VOI as a function of the resultant force direction, projected on the quadrant of the sphere shown in Fig. 3a. The values found are largely determined by the magnitude of the resultant force, which was maximal (200 N) for a resultant force working in the distal direction (at h"90°). 6(b). Projected contour plot of the standard deviation for the tissue principal strain in the VOI normalized by the average principal strain. The standard deviations in the yellow areas are relatively high, those in the red areas low. Since a low value indicates a more uniform strain distribution in the VOI, these are the preferred loading directions. direction. The lower values found for forces acting near the medial—superior side (indicated by the dark-red spot in Fig. 6b) indicate that loads acting in the lateral— inferior direction produce a more uniform strain distribution in the VOI. However, no single force direction produces uniform tissue strains. 4. Discussion The objective of this study was to estimate physiological in situ stresses and strains in trabecular bone tissue of a proximal canine femur. With the new techniques applied here it is now, for the first time, possible to obtain realistic estimates for in situ trabecular tissue stresses and strains. This accomplishment allowed us to quantitatively test the trajectorial hypothesis put forward by Wolff. It should be noted, however, that a number of assumptions and simplifications were made in this study that need some discussion. First, no muscle forces were applied to the model. Although muscle forces will add to the tissue stresses and strains in the trochanteric region and in the femoral shaft, the geometry of the femur is such that muscle forces alone (that is in the absence of joint forces) will not induce significant stresses or strains in the femoral head. It should be noted that the net effect of the muscle forces on the hip joint force is accounted for in the model, since the hip joint forces were obtained from in vivo measurements. Nevertheless, it is possible that muscle forces attaching to the proximal femur can slightly affect the local stress/strain calculation in the femoral head. Second, a unique Young’s modulus for all tissues was used in the model, whereas several investigators have found that the trabecular tissue properties are less than those of cortical bone (Rho et al., 1993; Choi et al., 1990; Kuhn et al., 1989). Hence, it is possible that the trabeculae are somewhat too stiff. Third, the modeling of trabecular bone with unique cubic elements produces ‘jagged’ surfaces. It has been shown that this can lead to errors in the stress-strain calculations, resulting in oscillating values, in particular at the bone surfaces (Jacobs et al., 1993; Guldberg and Hollister, 1994, Camacho et al., 1997). However, in an earlier study we found that the height of any of the bars in stress and strain histograms calculated from FE-models with an element size of 80 lm, differed by less than 7% from those calculated from FE-models representing the same structure with 20 lm elements (Van Rietbergen et al., 1995). Other researchers have found errors in the same range (1%—7%) for the 450 B. Van Rietbergen et al. / Journal of Biomechanics 32 (1999) 443—451 calculated average Von Mises stress when comparing results from digital based models with those from conventional smooth FE-models (Guldberg and Hollister, 1994). Consequently, although local oscillations and errors in the calculated stresses and strains might exist, their effect on the histograms and the calculated average and standard deviations will be small. We thus concluded that the model was sufficiently converged and that only a minor part of the wide variation in tissue stresses and strains found in this study can be due to numerical artifacts related to these jagged surfaces. Fourth, no cartilage was modeled and the cup, added for more realistic load transfer, was fixed to the femoral head. This implies that, in theory, shear stresses can be transferred as well. Since, however, the orthonormal forces were directed towards the center of the head, the interface shear stresses were small and the net shear stresses were zero. Fifth, the calculation of tissue stresses and strains by superimposing results does not account for the changes in contact area during flexion/extension of the femur. It is thus possible that the superimposed results are less accurate than those for the orthogonal load directions. Finally, no accurate information about the magnitude of the hipjoint force was available for this particular dog. The tissue strains calculated in the cortical region (958 lstrain) are larger than those measured from in vivo strain-gauge measurements during normal walking for the same region (325—502 lstrain; Page et al., 1993). These differences could not be explained by the fact that the strains in the model were not calculated exactly at the bone surface, since it would be expected that strains on the surface would be even slightly higher due to bending of the femur. It is possible, however, that the 200 N force chosen to represent the stance phase of loading is chosen somewhat too high, and better represents the forces during running or other more strenuous activities. Nevertheless, within the context of these limitations, it is possible to address the questions posed in the introduction. One of the questions was how large the range for the tissue stresses and strains is, and if this range can be estimated from average tissue values as determined from apparent-level FE models. It was found that the range for the tissue principal strains, as quantified by the standard deviations, depends on the force direction. The largest standard deviation for the principal strain distribution for any of the resultant force directions investigated was almost equal to the average tissue value. This indicates that for some 95% of the tissue in the VOI, the tissue stresses and strains are in a range defined by zero to three times their average tissue value. The smallest standard deviation was found for a force acting on the medial-superior quadrant in the region indicated by the red spot in Fig. 6b. For this force direction, the standard deviations were only half of the average value, indicating that with forces in this direction 95% of the tissue principal strains are in a range given by zero to twice their average tissue value. One of the most interesting results is that no single force creates uniform stress or strain distributions in the trabeculae. Even in the most optimal direction, the range of tissue stresses and strains is still rather large. This may seem to contradict the hypothesis put forward by Wolff. However, the hip joint force in the dog during a walking cycle varies in a large number of directions and even upward directed forces are possible during the swing phase (Bergmann et al., 1984; Bergmann, 1997). Hence, it is likely that the bone is adapted to withstand this total range of forces rather than one single force. The same conclusion was drawn in other studies based on the finding that non-orthogonal trabecular architectures are frequently found near many joints (Hert, 1992; Pidaparti and Turner, 1997). Since Wolff’s law, which was based upon assumptions drawn from unidirectional loading, can only explain the existence of orthogonal architectures, it was suggested in these studies that an optimal cancellous structure may appear differently under multidirectional joint loads than the ‘trajectorial’ organization proposed by Wolff (Pidaparti and Turner, 1997). The assumption that bone adapts to multiple load directions is also supported by the results of earlier load adaptive bone remodeling simulation studies, where it was found that multiple femoral loads are required to reproduce the natural bone density in a computer model of the femur (Carter et al., 1989; Weinans et al., 1992). It is thus possible that a more uniform distribution is found for tissue stresses and strains integrated over a whole walking cycle. Since the superimposing of loads is only possible within the quadrant indicated in Fig. 3b, this was not investigated in the present study. Although the range of tissue stresses and strains for the force representing the stance-phase is rather large, excessive values are found at few locations, and the maximum tissue stress found (60.2 MPa) is still less than values reported for the tissue fatigue strength (100—140 MPa) (Choi and Goldstein, 1992). These results indicate that trabecular bone has a safety factor of approximately two for forces that occur during normal activities. This value may seem rather low. As indicated before, however, the loads applied in the present analysis might be more representative for strenuous activities. Furthermore, it is possible that the micro damage generated in the bone is repaired by remodeling of the bone before fractures will occur. Finally, it should be noted that the calculated maximum tissue stress might be somewhat affected by the inaccuracies due to the ‘jagged’ surface. Jacobs et al. (1993) found that errors in the stresses calculated near the surface could be as large as 66%, although it was found in an earlier study (Guldberg and Hollister, 1994) that large errors are only found in regions of low stress. While recognizing these limitations, we think that the value found in the present study can be a reasonable first estimate for the safety factor of bone. B. Van Rietbergen et al. / Journal of Biomechanics 32 (1999) 443—451 Although the determination of tissue stresses and strains from large-scale FE-models has its limitations, it is shown in this study that this FE-approach can provide new information about the tissue loading conditions, that cannot be obtained by other methods. This new information can be of great importance for a better understanding of mechanically induced processes in bone. Acknowledgements This study was supported by NCF (Dutch National Computer Facilities) and a Cray Research University grant. References Bay, B.K., Hamel, A.J., Olson, S.A., Sharkey, N.A., 1997. Statistically equivalent load and support conditions produce different hip joint contact pressures and periacetabular strain. Journal of Biomechanics 30, 193—196. Beaupré, G.S., Hayes, W.C., 1985. Finite element analysis of a threedimensional open-celled model for trabecular bone. Journal of Biomechanical Engineering 107, 249—256. Bergmann, G., 1997. Personal communication. Bergmann, G., Siraky, J., Rohlmann, A., Koelbel, R., 1984. A comparison of hip joint forces in sheep, dog and man. Journal of Biomechanics 12, 907—921. Camacho, D.L., Hopper, R.H., Lin, G.M., Myers, B.S., 1997. An improved method for finite element mesh generation of geometrically complex structures with application to the skullbase. Journal of Biomechanics 30, 1067—1070. Carter, D.R., Orr, T.E., Fyhrie, D.P., 1989. Relationships between loading history and femoral cancellous bone architecture. Journal of Biomechanics 22, 231—244. Choi, K., Goldstein, S.A., 1992. A comparison of the fatigue behavior of human trabecular and cortical bone tissue. Journal of Biomechanics 25, 1371—1381. Choi, K., Kuhn, J.L., Ciarelli, M.J., Goldstein, S.A., 1990. The elastic moduli of human subchondral, trabecular, and cortical bone tissue and the size dependency of cortical bone modulus. Journal of Biomechanics 23, 1103—1113. Eisenhart-Rothe, R., Eckstein, F., Müller-Gerbl, M., Landgraf, J., Rock, C., Putz, R., 1997. Direct comparison of contact areas, contact stress and subchondral mineralization in human hip joint specimens. Anatomy and Embryology 195, 279—288. Feldkamp, L.A., Goldstein, S.A., Parfitt, A.M., Jesion, G., Kleerekoper, M., 1989. The direct examination of three dimensional bone architecture in vitro by computed tomography. Journal of Bone Mineral Research 4, 3—11. 451 Fyhrie, D.P., Hamid, M.S., Kuo, R.F., Lang, S.M., 1992. Direct threedimensional finite element analysis of human vertebral cancellous bone. Transactions of 38th Annual Meeting Orthopaedic Research Society, 551. Guldberg, R.E., Hollister, S.J., 1994. Finite element solution errors associated with digital image-based mesh generation. ASME/BED 28, 147—148. Hert, J., 1992. A new explanation of the cancellous bone architecture. Functional Development and Morphology, 2, 17—24. Hollister, S.J., Brennan, J.M., Kikuchi, N., 1993. A homogenization sampling procedure for calculating trabecular bone effective stiffness and tissue level stress. Journal of Biomechanics 26. Hollister, S.J., Goldstein, S.A., 1993. Trabecular tissue strain estimates for bone adapted to controlled implant loads. ASME/BED 24, 254—257. Hollister, S.J., Kikuchi, N., 1993. Comparison of trabecular tissue strains computed by homogenization theory to direct finite element calculations. ASME/BED 24, 258—261. Jacobs, C.R., Mandell, J.A., Beaupré, G.S., 1993. A comparative study of automatic finite element mesh generation techniques in orthopaedic biomechanics. ASME/BED 24, 512—514. Kuhn, J.L., Goldstein, S.A., Choi, K.W., London, M., Feldkamp, L.A., Matthews, L.S., 1989. Comparison of the trabecular and cortical tissue moduli from the human iliac crests. Journal of Orthopaedic Research 7, 876—884. Odgaard, A., Andersen, K., Ullerup, R., Frich, L.H., Melsen, F., 1994. Three-dimensional reconstruction of entire vertebral bodies. Bone 15, 335—342. Page, A.E., Allan, C., Jasty, M., Harrigan, T.P., Bragdon, C.R., Harris, W.H., 1993. Determination of loading parameters in the canine hip in vivo. Journal of Biomechanics 26, 571—579. Pidaparti, R.M., Turner, C.H., 1997. Cancellous bone architecture: advantages of nonorthogonal trabecular alignment under multidirectional joint loading. Journal of Biomechanics 30, 979—983. Rho, J.Y., Ashman, R.B., Turner, C.H., 1993. Young’s modulus of trabecular and cortical bone material: ultrasonic and microtensile measurements. Journal of Biomechanics 26, 111—119. Rüegsegger, P., Koller, B., Müller, R., 1996. A microtomographic system for the nondestructive evaluation of bone architecture. Calcificaion Tissue International 58, 24—29. Van Rietbergen, B., Weinans, H., Huiskes, R., Odgaard, A., 1995a. A new method to determine trabecular bone elastic properties and loading using micromechanical finite-element models. Journal of Biomechanics 28, 69—81. Van Rietbergen, B., Weinans, H., Polman, B.J.W., Huiskes, R., 1996. Computational strategies for iterative solutions of large FEM applications employing voxel data. International Journal of Numerical Methods in Engineering 39, 2743—2767. Weinans, H., Huiskes, R., Grootenboer, H.J., 1992. Effects of material properties of femoral hip components on bone remodeling. Journal of Orthopaedic Research 10, 845—853. Weinans, H., Huiskes, R., van Rietbergen, B., Sumner, D.R., Turner, T.M., Galante, J.O., 1993. Adaptive bone remodeling around bonded noncemented THA: a comparison between animal experiments and computer simulation. Journal of Orthopaedic Research 11, 500—513.